Associations of pregnancy complications and neonatal characteristics with bipolar disorder in the offspring: Nationwide cohort and sibling-controlled studies

Abstract

Objectives:

To investigate associations of neonatal characteristics and pregnancy complications with bipolar disorder (BPD) in offspring.

Methods:

We conducted a nationwide cohort study among 2,059,578 non-malformed singleton live-births in Sweden born 1983–2004. Using national registries with prospectively recorded information, we followed participants for a BPD diagnosis from 13 up to 34 years of age. We compared BPD risks between exposure categories using hazard ratios (HR) with 95% confidence intervals (CI) from adjusted Cox models. We also conducted sibling-controlled analyses among 1,467,819 full siblings.

Results:

There were 14,998 BPD diagnoses. Risk of BPD was 0.74% through 25 years of age. Very/extremely preterm birth (22 to 31 weeks) was related to increased BPD HRs in sibling-controlled analyses; compared with a gestational age of 37 weeks, adjusted HR (95% CI) for 31, 28, and 22 weeks were, respectively, 1.31 (0.99, 1.74), 2.09 (1.15, 3.79), and 5.74 (1.15, 28.63). Spontaneous but not medically indicated very/extremely preterm birth was associated with increased risk. Compared with vaginal birth, caesarean section birth was associated with 1.20 (1.08, 1.33) and 1.58 (1.06, 2.36) times higher BPD risk in general and sibling cohorts, respectively. Small-for-gestational age (SGA) birth was related to increased BPD HRs in general cohort and sibling analyses (HRs [95% CI] were 1.22 [1.06, 1.39] and 1.68 [1.13, 2.50], respectively); only term SGA was associated with increased risk. Head circumference-for-gestational age, gestational diabetes, preeclampsia, and placental abruption were not associated with BPD.

Conclusions:

Very/extremely preterm birth, caesarean birth, and SGA are related to BPD incidence.

INTRODUCTION

Bipolar disorder (BPD) affects nearly 40 million people worldwide, and is characterized by recurring cycles of manic and depressive episodes.¹ The main known risk factor for BPD is a family history of the disorder. Environmental stressors, including stressful life events or substance use, are also thought to contribute.² Current research seeks to identify potential modifiable risk factors of the disorder.

Some studies have suggested that pre- and perinatal characteristics, some of which could be modifiable, may be linked to BPD. However, overall evidence is inconsistent. Preterm birth was strongly positively associated with BPD in a population-based cohort study,³ as well as in two population-based case-control studies, but their statistical power was limited.^4,5 In addition, elective caesarean birth was associated with increased BPD risk in two population-based studies^5,6 but not in a smaller cohort study.⁷ Small-for-gestational age (SGA) birth was related to BPD in two population-based studies^8,9 but not in two other studies.^3,5 Small head circumference (<32 cm) has also been positively associated with BPD,¹⁰ although head circumference-for-gestational age has not been investigated. Further studies are necessary to determine whether neonatal characteristics are consistently associated with BPD. It is also unclear if pregnancy complications that may affect neonatal outcomes, such as gestational diabetes and preeclampsia, are related to risk of BPD. Prior studies on the topic are limited to those using obstetric complications scores, based on number and severity of experienced adverse pregnancy or perinatal conditions.^11,12 This approach limits the understanding of BPD etiology since offspring exposed to different individual risk factors with unrelated mechanisms may be assigned the same exposure level.

We used data from Swedish population registers to examine whether neonatal characteristics, including gestational age, mode of delivery, birth weight-for-gestational age, and head circumference-for-gestational age, are related to BPD risk in a nationwide cohort. We also assessed whether underlying causes of preterm birth or SGA, including gestational diabetes, preeclampsia, and placental abruption, are associated with BPD in the offspring. Finally, we performed nested sibling-controlled comparisons to account for potential confounding by time-invariant shared familial (genetic and environmental) factors and enhance causal inference.

METHODS

Study design

We conducted a population-based cohort study among live singleton children born at ≥22 completed gestational weeks between 1983 and 2004, who were recorded in the Swedish Medical Birth Register. The National Board of Health and Welfare and Statistics Sweden provided information from nationwide registers. Information in the Birth Register¹³ was cross-linked with the National Patient-,¹⁴ Total Population-,¹⁵ Education-,¹⁶ Multi-generation-,¹⁷ and Cause of Death Registers using the person-unique national registration number assigned to all Sweden residents at birth or immigration.¹⁸ The Birth Register includes information on prenatal, obstetric, and neonatal care for more than 98% of all births in Sweden. The National Patient Register includes diagnoses at discharge from hospital admissions since 1987 and from outpatient hospital visits since 2001. Diagnoses are coded according to the Swedish version of the International Classification of Diseases (ICD); eighth revision (ICD-8) between 1985 and 1986, ninth revision (ICD-9) between 1987 and 1996, and tenth revision (ICD-10) since 1997. The study was approved by the Regional Ethical Review Board in Stockholm, Sweden (No. 2018/5:2). Informed consent was not required.

Outcome

BPD was defined as a main or secondary diagnosis of ICD-9 (296A-296E, 296W, or 296X) or ICD-10 (F30 or F31) codes recorded at least twice on different occasions in the National Patient Register or Cause of Death Register starting at 13 years of age; this was based on a validated algorithm.¹⁹ Age at BPD diagnosis was recorded as the age when the first diagnosis appeared. ICD codes for all diagnoses are presented in Table S1.

Exposures

The primary exposures of interest were neonatal characteristics, including gestational age at birth, birth weight-for-gestational age, head circumference-for-gestational age, and mode of delivery. Secondary exposures were pregnancy complications that may cause preterm birth or SGA, or that may affect the mode of delivery; these included gestational diabetes, preeclampsia, and placental abruption. Gestational age was obtained by using the following hierarchy: early second trimester ultrasound (65.3%), date of the last menstrual period (31.8%), or a postnatal assessment (3.0%). Births were classified as post-term (≥42 completed weeks), term (37 to 41 weeks), moderately preterm (32 to 36 weeks), very preterm (28 to 31 weeks), or extremely preterm (22 to 27 weeks). Preterm birth was further classified as spontaneous or medically indicated (i.e., induced), as recorded in the Birth Register. Birth weight-for-gestational age was defined using the ultrasound-based Swedish reference for fetal growth,²⁰ and SGA was defined as a birth weight-for-gestational age <3^rd percentile. Head circumference-for-gestational age was defined using Swedish reference standards,²¹ and small head circumference (SHC) was defined as a head circumference-for-gestational age <3^rd percentile. SGA and SHC were further classified as term (≥37 weeks) or preterm (<37 weeks). Mode of delivery was classified as non-instrumental vaginal, instrumental vaginal, elective caesarean section, or emergency caesarean section; this information was available in the Birth Register starting in 1990. Information on pregnancy complications was obtained from the Birth Register according to the ICD codes presented in Table S1.

Covariates

Covariate information was primarily extracted from the Birth Register, but also from the Total Population and Education Registers. Maternal age at delivery was the date of delivery minus the mother’s birth date. Mother’s country of birth (from the Total Population Register) was categorized as Nordic vs. non-Nordic. Maternal education was the lifetime highest level of completed education. Information on whether the mother cohabited with the child’s father was obtained at the first prenatal visit. Parity was the number of births. Maternal height was self-reported at the first prenatal visit; for multiparous women, we took the median height across pregnancies to decrease error.²² Early pregnancy body mass index (BMI, kg/m²) was calculated from height and weight measured objectively in light clothing at the first prenatal visit, which in Sweden occurs before 14 weeks of gestation in 90%.²³ BMI was classified as underweight (BMI <18.5), normal weight (18.5–24.9), overweight (25.0–29.9), obesity grade 1 (30.0–34.9), obesity grade 2 (35.0–39.9), or obesity grade 3 (≥40.0).²⁴ Smoking was determined by self-report at either the first prenatal visit or in the third trimester; this has been validated with cotinine markers.²⁵ Maternal infection was defined according to the ICD codes presented in Table S1. Paternal age was the date of delivery minus the father’s birth date. Parental BPD was defined as at least one diagnosis in the National Patient Register. Parental psychiatric disorder other than BPD was the presence of at least one of the following diagnoses in the National Patient Register: non-affective psychosis, depression, anxiety, substance abuse, personality disorder, attention-deficit/hyperactivity disorder, or autism spectrum disorder. Diagnoses of parental psychiatric disorders were defined as at least one ICD-8, ICD-9, or ICD-10 code, as presented in Table S1.

Statistical Analysis

General cohort analyses.

The general cohort comprised children born January 1983 through December 2004 who were followed starting at age 13 years until the earliest of two BPD diagnoses, emigration, death, or December 31^st, 2017.

We estimated BPD risk by age 25 years, the mean age at diagnosis,²⁶ using the Kaplan Meier method and compared BPD risks by categories of exposures using adjusted hazard ratios (HR) with 95% confidence intervals (CI) from Cox proportional hazards models. The robust sandwich estimate of the covariance matrix was used to compute 95% CI, to account for the correlation of measures among women with more than one pregnancy in the dataset. Models were adjusted for independent predictors of BPD that could be related to exposures without being their consequences, per prior knowledge. These included maternal age, country of origin, cohabitation with the child’s father, education level, parity, height, early-pregnancy BMI, smoking during pregnancy, maternal infection, presence of a diagnosis of BPD²⁷ or other psychiatric disorder in the mother or the father, paternal age, and child sex and year of birth (Figure S1). Models with neonatal characteristics as the primary exposures were additionally adjusted for pregnancy complications. Models with gestational diabetes and placental abruption as exposures were also adjusted for preeclampsia, and the model with preeclampsia as the exposure was additionally adjusted for gestational diabetes. We also considered gestational age as a continuous exposure and examined potential non-linear associations with BPD with use of Cox models with restricted cubic splines. Knots were placed at 28, 32, 37, and 42 weeks of gestational age.

Preterm birth has been related to increased incidence of BPD³ and could be a consequence of SGA; hence, an association between SGA and BPD^8,9 could be mostly driven by preterm birth. We evaluated the role of SGA independent of preterm birth by estimating the proportion of its association with BPD that was not mediated through gestational age, using causal mediation analyses under the assumptions of a potential outcomes frame, detailed elsewhere.^28,29

Sibling cohort analyses.

Examining the associations of neonatal characteristics and pregnancy complications with BPD among siblings offers an opportunity to enhance causal inference by controlling for time-invariant shared confounding factors. Full siblings share up to one-half of autosomal DNA; thus, confounding by unmeasured genetic characteristics is less likely in studies comparing siblings with each other than in comparisons of unrelated individuals. Full sibling comparisons also control for environmental factors shared by siblings, and for genetic and all other time-invariant characteristics of the parents. We identified full siblings in the general cohort with use of the Multigeneration Register and assembled a sibling cohort consisting of children with at least one full sibling in the general cohort. We conducted sibling comparisons by estimating HR with 95% CI through stratified Cox models in which each family was a stratum. Models were adjusted for child birth order and sex, maternal BMI, and smoking during pregnancy. The model with mode of delivery as the exposure was additionally adjusted for preeclampsia and placental abruption. For continuous exposures, we also compared BPD risks with use of stratified Cox proportional hazards models with restricted cubic splines;³⁰ these models were adjusted for birth order and child sex.

We noted that children included in the sibling cohort differed from those who were excluded with respect to outcome incidence, exposure prevalence, and sociodemographic characteristics distributions. Compared with children in the sibling cohort, those excluded had higher incidence of the outcome, higher prevalence of the exposures, less favorable sociodemographic conditions, and higher parental prevalence of BPD and other psychiatric disorders (Table S2). The reason for exclusion from the sibling cohort was lacking full siblings in the Birth Register during the birth years that defined the general cohort. This could be due to having older siblings born before systematic follow-up through the Patient Registers could be accomplished, or to not having any live siblings. Because selection into the sibling cohort could bias the estimates of association, we corrected the stratified estimates for potential selection and confounding biases via inverse probability weighting (IPW). We calculated HR with 95% CI from Cox regression models with stabilized weights, which were the product of the inverse of the probability of exposure as a function of measured covariates times the inverse of the probability of being selected into the sibling cohort as a function of covariates.³¹ Analyses were conducted with use of SAS version 9.4 (SAS Institute).

RESULTS

From January 1983 through December 2004, the Birth Register included 2,176,987 live singleton births. We excluded those with missing maternal (n=14,228) and child (n=2885) national registration numbers, children who emigrated (n=69,908) or died (n=12,639) before 13 years of age, and children with congenital malformations (n=17,749). Hence, the general cohort comprised 2,059,578 children with 14,998 BPD diagnoses over a median 24.5 years of age (interquartile range [IQR] 19.0, 29.0). The risk of BPD was 0.74% through 25 years of age. BPD was comorbid with anxiety, attention-deficit/hyperactivity disorder, personality disorder, substance abuse, and eating disorders in 64%, 33%, 24%, 18%, and 12% of cases, respectively (Table S3). BPD risk increased with maternal age <25 and ≥35 years, Nordic country of birth, education (≥15 years), parents not cohabiting, BMI, smoking and infections during pregnancy, paternal age <25 and ≥40 years, maternal and paternal BPD or other psychiatric disorder, offspring’s female sex, and neonatal sepsis; and decreased with parity (Table S4). The sibling cohort included 1,467,819 full siblings from 640,321 families with 10,008 BPD cases over a median 24.5 years of age (IQR 19.8, 28.6).

In both the general and sibling cohorts, gestational diabetes, placental abruption, gestational age ≤31 weeks, and SGA were related to increased risk of BPD in unadjusted analyses (Table 1). BPD risk was increased for both spontaneous and medically indicated preterm birth; the association was strongest for spontaneous very/extremely preterm birth. Birth by elective or emergency caesarean section was associated with an increased risk of BPD, compared with non-instrumental vaginal birth. Head circumference-for-gestational age was generally unrelated to BPD risk.

Table 1.

Pregnancy, delivery, and neonatal characteristics and incidence of bipolar disorder (BPD)^a in offspring. Live-born singleton non-malformed children in Sweden 1983–2004.

	General cohort			Sibling cohort
	Number of children	No. with BPD	Riskb by age 25 years, %	Number of children	No. with BPD	Riskb by age 25 years, %
Total	2,059,578	14,998	0.74	1,467,819	10,008	0.70
Pregnancy complications
Gestational diabetesc
No	2,039,173	14,867	0.74	1,453,694	9918	0.70
Yes	13,865	93	1.04	9770	63	0.98
Missing	6540	38	0.91	4355	27	0.88
Preeclampsia
No	2,004,810	14,642	0.74	1,432,545	9790	0.70
Yes	54,768	356	0.71	35,274	218	0.65
Placental abruption
No	2,050,446	14,925	0.74	1,461,868	9968	0.70
Yes	9132	73	0.79	5951	40	0.79
Neonatal characteristics
Gestational age at birth (weeks)
Post-term (≥42)	165,640	1195	0.72	113,959	776	0.65
Term (37 to 41)	1,787,969	12,971	0.74	1,285,693	8712	0.70
Moderately preterm (32 to 36)	91,458	704	0.75	59,863	440	0.71
Very preterm (28 to 31)	8844	72	0.78	5159	50	0.92
Extremely preterm (22 to 27)	2706	28	1.38	1498	17	1.33
Missing	2961	28	0.59	1647	13	0.26
Type of birth
Term or post-term (≥37 weeks)	1,953,609	14,166	0.74	1,399,652	9488	0.70
Moderately preterm, spontaneous	42,108	203	0.89	28,307	131	0.76
Very/extremely preterm, spontaneous	4614	40	1.41	2798	27	1.38
Moderately preterm, medically indicated	13,448	68	0.90	8707	48	0.90
Very/extremely preterm, medically indicated	2557	10	0.70	1416	7	0.70
Missing	43,242	511	0.62	26,939	307	0.60
Mode of deliveryd
Vaginal non-instrumental	1,041,481	4888	0.87	789,195	3633	0.79
Vaginal instrumental	88,536	379	0.91	53,077	227	0.77
Elective caesarean section	77,461	374	1.00	53,366	258	0.92
Emergency caesarean section	79,022	393	1.08	46,950	247	1.03
Missing	773,078	8964	0.58	525,231	5643	0.57
Birth weight-for-gestational age, percentiles
<3	38,842	360	0.88	23,827	201	0.79
3 to <10	104,778	846	0.74	68,184	528	0.70
10 to 90	1,656,547	11,974	0.74	1,183,523	8022	0.70
>90 to 97	173,779	1233	0.77	130,212	871	0.74
>97	73,591	477	0.69	54,759	331	0.69
Missing	12,041	108	0.75	7314	55	0.65
Small-for-gestational age (SGA)e by gestational age
No SGA (≥37 weeks)	1,916,226	13,808	0.74	1,376,245	9287	0.70
SGA (≥37 weeks)	30,809	304	0.94	19,264	175	0.83
No SGA (<37 weeks)	92,469	722	0.78	60,433	465	0.74
SGA (<37 weeks)	8033	56	0.64	4563	26	0.60
Missing	12,041	108	0.75	7314	55	0.65
Head circumference-for-gestational age, percentiles
<3	46,613	422	0.68	31,792	283	0.67
3 to <10	141,396	1255	0.79	97,811	812	0.71
10 to 90	1,602,710	11,647	0.74	1,149,004	7807	0.70
>90 to 97	134,089	860	0.70	95,586	588	0.68
>97	52,531	333	0.71	36,638	214	0.70
Missing	82,239	481	0.81	56,988	304	0.73
Small head circumference-for-gestational age (SHC)f by gestational age
No SHC (≥37 weeks)	1,847,211	13,437	0.74	1,324,706	9005	0.70
SHC (≥37 weeks)	40,917	375	0.68	28,233	256	0.67
No SHC (<37 weeks)	83,515	658	0.76	54,333	416	0.73
SHC (<37 weeks)	5696	47	0.69	3559	27	0.68
Missing	82,239	481	0.81	56,988	304	0.73

^aBPD starting at 13 years of age or later.

^bEstimated using the Kaplan Meier method.

^cExcludes women with pregestational diabetes.

^dData available since 1990.

^eBirth weight-for-gestational age percentile <3.

^fHead circumference-for-gestational age percentile <3.

Gestational diabetes, preeclampsia, and placental abruption were not associated with adjusted HR of BPD in cohort or sibling-controlled analyses (Table 2). Among very/extremely preterm births, adjusted BPD HR increased with decreasing gestational age in both cohort and sibling-controlled analyses (Figure 1A, Table S5). In the sibling cohort, compared with a gestational age of 37 weeks, adjusted HR (95% CI) for 31, 28, and 22 weeks were 1.31 (0.99, 1.74), 2.09 (1.15, 3.79), and 5.74 (1.15, 28.63), respectively. Spontaneous but not medically indicated very/extremely preterm birth was associated with increased HR of BPD (Table 2); compared with term birth, the adjusted HR (95% CI) was 2.06 (1.38, 3.09) in the general cohort. A similar association was observed among siblings, but statistical power was limited (Figure 1B, Table S5).

Table 2.

Pregnancy, delivery, and neonatal characteristics and hazard ratios for bipolar disorder (BPD)^a in the general and sibling cohorts. Live-born singleton non-malformed children in Sweden 1983–2004.

	General Cohortb	Sibling Cohortc
	Adjusted hazard ratio (95% CI)d,e	Adjusted hazard ratio (95% CI)f	IPW-adjusted hazard ratio (95% CI)g
Pregnancy complications
Gestational diabetes
No	1.00	1.00	1.00
Yes	1.04 (0.78, 1.39)	1.29 (0.62, 2.70)	1.22 (0.56, 2.68)
Preeclampsia
No	1.00	1.00	1.00
Yes	0.92 (0.80, 1.05)	1.13 (0.79, 1.62)	1.09 (0.75, 1.57)
Placental abruption
No	1.00	1.00	1.00
Yes	1.06 (0.76, 1.47)	0.75 (0.33, 1.69)	1.14 (0.51, 2.55)
Neonatal characteristics
Gestational age at birth (weeks)
Post-term (≥42)	0.98 (0.91, 1.06)	0.92 (0.76, 1.11)	0.93 (0.77, 1.13)
Term (37 to 41)	1.00	1.00	1.00
Preterm (22 to 36 weeks)	1.03 (0.94, 1.14)	1.05 (0.80, 1.38)	1.02 (0.78, 1.34)
Type of birth
Term or post-term (≥37 weeks)	1.00	1.00	1.00
Moderately preterm, spontaneous	0.95 (0.79, 1.15)	0.87 (0.52, 1.45)	1.04 (0.58, 1.88)
Very/extremely preterm, spontaneous	2.06 (1.38, 3.09)	2.18 (0.81, 5.89)	1.70 (0.62, 4.65)
Moderately preterm medically, indicated	1.07 (0.78, 1.48)	1.43 (0.68, 3.00)	1.65 (0.71, 3.82)
Very/extremely preterm medically, indicated	0.64 (0.24, 1.71)	1.81 (0.16, 21.0)	-
Mode of deliveryh
Vaginal non-instrumental	1.00	1.00	1.00
Vaginal instrumental	1.08 (0.94, 1.23)	1.14 (0.76, 1.71)	1.14 (0.73, 1.79)
Any caesarean section	1.20 (1.08, 1.33)	1.21 (0.81, 1.81)	1.58 (1.06, 2.36)
Elective caesarean section	1.10 (0.95, 1.28)	1.16 (0.69, 1.95)	1.54 (0.90, 2.64)
Emergency caesarean section	1.29 (1.13, 1.48)	1.24 (0.78, 1.98)	1.53 (0.96, 2.44)
Birth weight-for-gestational age, percentiles
<3	1.22 (1.06, 1.39)	1.68 (1.13, 2.50)	1.64 (1.09, 2.47)
3 to <10	1.01 (0.92, 1.11)	1.18 (0.94, 1.49)	1.17 (0.93, 1.47)
10 to 90	1.00	1.00	1.00
>90 to 97	1.08 (1.00, 1.17)	1.10 (0.91, 1.33)	1.25 (1.04, 1.52)
>97	0.94 (0.83, 1.06)	1.03 (0.77, 1.38)	0.91 (0.66, 1.26)
Small-for-gestational age (SGA)i by gestational age
No SGA (≥37 weeks)	1.00	1.00	1.00
SGA (≥37 weeks)	1.28 (1.11, 1.47)	1.66 (1.09, 2.52)	1.61 (1.04, 2.50)
No SGA (<37 weeks)	1.05 (0.95, 1.16)	1.00 (0.75, 1.33)	0.97 (0.73, 1.29)
SGA (<37 weeks)	0.90 (0.62, 1.32)	1.40 (0.50, 3.90)	1.52 (0.46, 4.96)
Head circumference-for-gestational age, percentiles
<3	1.16 (1.02, 1.32)	0.95 (0.69, 1.30)	0.81 (0.59, 1.12)
3 to <10	1.09 (1.01, 1.17)	0.95 (0.78, 1.15)	0.89 (0.73, 1.08)
10 to 90	1.00	1.00	1.00
>90 to 97	1.05 (0.96, 1.15)	1.09 (0.88, 1.36)	0.95 (0.76, 1.17)
>97	1.10 (0.96, 1.26)	1.05 (0.74, 1.49)	1.02 (0.72, 1.46)
Small head circumference-for-gestational age (SHC)j by gestational age
No SHC (≥37 weeks)	1.00	1.00	1.00
SHC (≥37 weeks)	1.14 (1.00, 1.31)	0.99 (0.72, 1.37)	1.00 (0.73, 1.37)
No SHC (<37 weeks)	1.01 (0.91, 1.13)	1.08 (0.80, 1.47)	1.02 (0.75, 1.39)
SHC (<37 weeks)	1.16 (0.81, 1.66)	0.54 (0.16, 1.88)	0.41 (0.10, 1.70)

^aBPD starting at 13 years of age or later.

^bThe cohort comprises 2,059,578 children with 14,998 cases of BPD.

^cThe cohort comprises 1,467,819 full siblings distributed in 640,321 families. There were 10,008 cases of BPD.

^dFrom proportional hazards models with age at first diagnosis of BPD as the outcome and each perinatal characteristic as the exposure. Models were adjusted for maternal age, country of origin, cohabitation with the child’s parent, education level, parity, height, early-pregnancy body mass index, smoking during pregnancy, preeclampsia, gestational diabetes, and infection, presence of a diagnosis of BPD, non-affective psychosis, depression, anxiety, substance abuse, personality disorder, attention-deficit/hyperactivity disorder, and autism spectrum disorder in the mother or the father, paternal age, and child sex and year of birth. The model for mode of delivery was additionally adjusted for placental abruption. A robust estimate of the variance was specified in all models to account for siblings.

^eComplete case analysis; n = 1,405,117 with 9,019 cases of BPD.

^fFrom proportional hazards models with age at first diagnosis of BPD as the outcome, stratified by family. Models were adjusted for birth order, early-pregnancy body mass index, smoking during pregnancy, and child sex. The model for mode of delivery was additionally adjusted for placental abruption. Complete case analyses; n = 831,667 with 4683 cases of BPD.

^gInverse probability weighting. Estimates are from weighted proportional hazards models. Stabilized weights were computed as the product of the inverse of exposure probability given the covariates in footnote 3 times the inverse of the probability of inclusion into the sibling cohort given covariates.

^hData available since 1990.

ⁱBirth weight-for-gestational age percentile <3.

^jHead circumference-for-gestational age percentile <3.

Figure 1. Gestational age at birth and bipolar disorder in live-born singleton non-malformed children in Sweden, 1983–2004. A. All births. B. Spontaneous births.

The general cohort included 2,059,578 children and the sibling cohort included 1,467,819 full siblings from 640,321 families. There were 14,998 and 10,008 BPD cases in the general and sibling cohorts, respectively. Hazard ratios (solid blue lines) and 95% CI (dashed blue lines) for BPD by gestational age were modeled with restricted cubic splines. Four knots were placed at 28, 32, 37, and 42 weeks gestational age. Estimates for the general cohort are from Cox proportional hazards models with maternal age, country of origin, cohabitation with the child’s parent, education level, parity, height, early-pregnancy body mass index, smoking during pregnancy, preeclampsia, gestational diabetes, and infection, presence of a diagnosis of BPD, non-affective psychosis, depression, anxiety, substance abuse, personality disorder, attention-deficit/hyperactivity disorder, and autism spectrum disorder in the mother or the father, paternal age, child sex and year of birth, and linear and spline terms for gestational age. Estimates for the sibling cohort are from Cox proportional hazards models with birth order, child sex, and linear and spline terms for gestational age, stratified by family.

In the general cohort, birth by any caesarean section, and specifically emergency caesarean section, was associated with increased HR of BPD (Table 2). The association of any caesarean section was also observed in IPW-adjusted sibling-controlled analyses; compared with non-instrumental vaginal births, any caesarean section was related to 1.58 times (95% CI: 1.06, 2.36) higher HR of BPD.

SGA was related to increased BPD HR in both the general and sibling cohorts (Table 2). This association was stronger in sibling-controlled analyses; compared with birth weight-for-gestational age >10^th and <90^th percentiles, SGA (<3^rd percentile) HR was 1.68 times (95% CI: 1.13, 2.50) higher. The SGA-related increase in HR was restricted to term SGA (Table 2). In mediation analysis, 99% of the association between (all) SGA and BPD was independent of preterm birth (Table 3). Head circumference-for-gestational age <3^rd percentile was associated with increased BPD HR in the general cohort, but not in sibling comparisons (Table 2).

Table 3.

Proportion of the association of small-for-gestational age (SGA) with bipolar disorder that is not mediated through preterm birth (gestational age at birth <37 weeks)

Complication	Hazard ratio (95% CI)a			% not mediated through preterm birth	P complication x preterm birth interaction
	Total	Indirect through preterm birth	Direct or indirect not through preterm birth
SGA	1.21 (1.06, 1.38)	1.00 (0.99, 1.01)	1.21 (1.06, 1.38)	99	0.06

^aFrom proportional hazards models with age at first diagnosis of bipolar disorder as the outcome adjusted for maternal age, country of origin, cohabitation with the child’s parent, education level, parity, height, early-pregnancy body mass index, smoking during pregnancy, infection, preeclampsia, placental abruption, and gestational diabetes, presence of a diagnosis of bipolar disorder, non-affective psychosis, depression, anxiety, substance abuse, personality disorder, attention-deficit/hyperactivity disorder, and autism spectrum disorder in the mother or the father, paternal age, and child sex and year of birth. The association between SGA and preterm birth was modeled with use of logistic regression.

DISCUSSION

In this nationwide cohort study of over 2.1 million people, very/extremely spontaneous preterm birth, caesarean section delivery, and term SGA were each associated with increased risk of BPD. Head circumference-for-gestational age was not associated with BPD incidence, nor were pregnancy complications, including gestational diabetes, preeclampsia, and placental abruption.

We found that BPD risk increased with decreasing gestational age among very/extremely preterm births. Previous studies have suggested an association between preterm birth and BPD, but did not distinguish between spontaneous or medically indicated preterm birth.^3–5 In a register-based cohort study of Sweden residents born in 1973 to 1985, those born very/extremely and moderately preterm had a 7.4- and 2.7-fold increased rate of BPD, respectively, compared with those born at term.³ In our study, the association was restricted to very/extremely preterm birth and was weaker in magnitude; this may be due to the use of different BPD case definitions. Our case definition required two diagnoses of BPD, but encompassed a wider set of ICD codes than the other study, potentially leading to the inclusion of a different set of cases. We also accounted for confounding by time-invariant shared familial factors in sibling-controlled analyses.

Our finding that only spontaneous very/extremely preterm birth was associated with increased BPD risk is novel. Differences in the associations by preterm birth types suggest that factors underlying spontaneous but not medically indicated preterm birth may be on the causal pathway to BPD. Very/extremely spontaneous preterm birth may represent defective placentation, whereas moderately preterm birth is more related to chorioamnionitis.³² One may only speculate if placental insufficiency may be more strongly related to long-term neurodevelopmental outcomes than chorioamnionitis.³³ On the other hand, medically indicated preterm birth often stems from fetal health concerns following preeclampsia, gestational diabetes, SGA, or other complications.³⁴ Pregnancy complications were not related to BPD in our study, and there was little indirect effect of SGA on BPD through preterm birth in mediation analyses. This provides internal consistency to the lack of association with medically indicated preterm birth, but we cannot rule out limited statistical power as an explanation for these findings.

We found that birth by any caesarean section was related to increased risk of BPD in cohort and IPW-adjusted sibling-controlled analyses. Two previous studies found an increased risk associated with elective caesarean section in the cohort analysis but not in the sibling comparison.^5,6 No association with emergency caesarean section was observed in these studies.^5,6 Still, possible explanations for an association between caesarean delivery and BPD include disrupted microbial colonization and altered immune system development. Infants born by caesarean section are not (elective caesarean) or less (emergency caesarean) exposed to the vaginal microbiota of the mother during birth, and have, compared with infants born vaginally, lower diversity of the gut microbiota through one to two years of age.³⁵ In addition, infants born by caesarean section are not exposed to the immune processes activated during labor, which in combination with differing microbial colonization, may impact the development of the immune system.³⁵ Immune mechanisms and gut microbiota have been implicated in BPD and other neuropsychiatric disorders.^36–40 However, we cannot completely rule out residual confounding by time-varying causes of caesarean section. Still, in our analyses the association remained after controlling for pregnancy complications that may indicate the need for a caesarean section, and these complications were generally unrelated to BPD risk. Further research on this topic is warranted given that rates of caesarean section delivery are increasing, and many may be unnecessary.⁴¹

We examined two measures of fetal growth in relation to BPD, birth weight-for-gestational age and head circumference-for-gestational age. We found that SGA was related to an increased risk of BPD in both the general and sibling cohorts. This risk increase was restricted to term SGA, a finding which was supported by mediation analyses showing that preterm birth did not mediate the SGA-BPD association. This is in contrast to a population-based Danish cohort study, finding that birth weight-for-gestational age <10^th percentile was associated with BPD among preterm children but not in term children.⁸ Other studies found no association with birth weight-for-gestational age ≤2 SD below the mean.^3,5 Varying definitions of SGA and the inclusion of different covariates may contribute to the different results across studies; some studies also had small cell counts. Risk factors for SGA differ by gestational age;⁴² while preterm SGA is strongly associated with preeclampsia, which was not associated with BPD in our study, specific risk factors for term SGA are less clear.⁴² A previous study showed that head circumference at birth <32cm was positively associated with BPD.¹⁰ When accounting for gestational age, we also found that small head circumference was associated with an increased risk of BPD in the general cohort. However, the association disappeared in the sibling comparison, which suggests that shared familial factors may have accounted for this association.

No prior investigations had examined pregnancy complications individually in relation to BPD in the offspring. We did not observe associations with gestational diabetes, preeclampsia, or placental abruption examined as individual risk factors. Of note, these conditions are causes of medically indicated preterm birth and this type of preterm birth was generally not related to BPD risk; this supports the notion that the risk increase in BPD associated with preterm birth is mostly due to the spontaneous type.

This study has several strengths. First, we used a validated algorithm to define BPD.¹⁹ Second, the population-based design with use of complete nationwide registers limits the possibility of selection bias since loss to follow-up for reasons other than emigration is minimized. Third, confounding by access to care and socioeconomic factors should be limited by the existence of universal, standardized healthcare in Sweden and the relative sociodemographic homogeneity of the Swedish population. Fourth, the validity of exposure variables from the Swedish Birth Register is excellent.¹³ Finally, examining the associations of neonatal characteristics and pregnancy complications with psychiatric outcomes among siblings offers an opportunity to enhance causal inference by controlling for time-invariant shared confounding factors. Full siblings share up to one-half of autosomal DNA; thus, confounding by unmeasured genetic characteristics is less likely in studies comparing siblings with each other than in comparisons of unrelated individuals. Full sibling comparisons also control for environmental factors shared by siblings, including early-life socioeconomic status, and for genetic and all other time-invariant characteristics of the parents.

There are also some limitations. First, the relative sociodemographic homogeneity of the Swedish population may limit the generalizability of the study findings to populations with different sociodemographic structures. Second, we were unable to distinguish between bipolar disorder type 1 and type 2. Third, the sibling cohort differed from the general cohort on outcome and exposure distributions, which could lead to selection bias; this was addressed by implementing IPW. Finally, the sibling comparison design is prone to some biases. Random measurement error in exposure⁴³ and inherent adjustment for potential mediators shared within families⁴⁴ could spuriously attenuate the sibling–controlled estimates. In addition, although sibling-controlled analyses enhance adjustment for confounders shared within families, they can produce biased estimates when non-shared confounders differ more among siblings than the exposure does,⁴³ even when adjustment is performed.⁴⁵ Furthermore, sibling comparison designs assume that the exposure or outcome status of one child does not affect the exposure or outcome of a sibling; the presence of potential exposure-to-exposure carryover effects are the most likely violation of this assumption in this case, and this type of carryover effect does not result in bias in the sibling comparison effect estimates.⁴⁶

In conclusion, very/extremely spontaneous preterm birth, caesarean birth, and SGA were each associated with increased risk of BPD in both cohort and sibling-controlled analyses. Head circumference-for-gestational age, gestational diabetes, preeclampsia, and placental abruption were not associated with BPD.

Data Availability:

The data underlying this article were obtained from the Swedish National Board of Health and Welfare and Statistics Sweden and cannot be shared publicly according to Swedish law. Researchers can apply for access to these data through the Swedish National Board of Health and Welfare and Statistics Sweden after obtaining an ethics approval from a regional ethical review board.